Geautomatiseerde segmentatie van druiventrossen in het verpakkingsproces


Automated segmentation of grape stem: packaging process and high-throughput phenotyping applications

Gepubliceerd op
22 september 2015

Computer vision and robotics part of the Wageningen UR Greenhouse technology group aims at automating processes in agriculture. As a consequence, the accuracy and the speed of the necessary tasks rapidly increases. Almost every filed related to food industry and agriculture benefits from automation of the processes within. More specifically, the work conducted on automated grapes and grapes parts segmentation is a part of the European PickNPack project that targets flexible automation of food packaging and accurate quality assessment of products using robots and computer vision techniques.


One typical task of the robot in this process is to pick a grape bunch from a crate and place it in a packaging box. In order for this task to be completed, the robot needs to ‘see’ the stem of the individual grape bunch, grab it using the thickest part of the stem location, and gently place it in the packaging box – similar to the way humans do it. We are able to accurately detect grapes stem using RGB colour images. Since the colours of stem and berries are often very similar, we use curvature information of the image next to the colour information. Some of the automated stem segmentations are shown in the image. This software is also applicable for automated grape parts segmentation (stem, leaves and berries) which is an essential part of the high-throughput phenotyping process.